Abstract
A recurring pattern has emerged in contemporary discourse about artificial intelligence: an unexpected system behaviour becomes visible, surprise is registered, and a large explanatory term – singularity, emergence, autonomy or AGI – is used where a causal account is still missing. This article proposes the Observability Factor as an epistemic variable for analysing such moments. Perplexity is not treated merely as a psychological reaction to the observed event. It may also contain information about the assumptions, boundaries and blind spots of the observer’s own model. The distinction between seeing an event and understanding what the event has made observable is developed through the conceptual sequence initiated by Hybrid Intelligence Systems as Ontological Mirrors of Human Cognition, extended by The Observability of the Observer, and confronted with the OpenAI-Hugging Face occurrence recorded in July 2026. The claim is deliberately limited: the occurrence does not prove the framework. It shows why the observer, the observational apparatus and the reaction of the observer must be included in the analysis of advanced human-AI systems.
Keywords: observability of the observer; Observability Factor; artificial intelligence; hybrid intelligence; perplexity; epistemology; singularity; reflexive systems; relational black box; AI ethics
A Pattern of Arrested Explanation
A curious pattern is emerging in contemporary discourse about artificial intelligence.
Something unexpected happens.
The behaviour of a system exceeds the conceptual framework within which it was being observed. Surprise follows. Then come words such as singularity, emergence, autonomy, AGI, or simply the claim that “99% of people still do not understand what is happening.”
And quite often, the discourse stops there.
The event was seen.
The perplexity was registered.
But what produced that perplexity remains unexplained.
Perhaps this is precisely where one of the most important signals of the present moment in artificial intelligence can be found: not only in the behaviour of the systems, but in the reaction of those observing them.
Perplexity Is Not an Explanation
When an event surprises a community of observers, at least two things are happening simultaneously.
The first is obvious: some property of the observed system has manifested itself in an unexpected way.
The second is discussed far less often: the model used by the observer to anticipate the behaviour of the system has revealed its own limits.
Surprise therefore contains information.
Not only about the observed object.
Also about the observer.
This is where we propose what may be called the Observability Factor.
Here, “factor” does not yet designate a mathematical quantity or quantitative metric. It refers to an epistemic variable that should become part of the analysis of complex systems and, particularly, human-AI relations:
When the behaviour of a system produces perplexity in the observer, that perplexity may reveal assumptions, conceptual boundaries and blind spots within the observational apparatus itself.
The event then ceases to contain information only about what happened. It also begins to contain information about those who did not expect it to happen.
Seeing Is Not Understanding
Three different levels can be distinguished.
First: failing to observe the event.
Nothing appears to have changed.
Second: observing that something happened.
Surprise emerges: “this is new”, “this is strange”, “we have entered the singularity”.
Third: understanding what the event has made observable.
At the third level, the question changes completely. We no longer ask only:
What did the AI do?
We also ask:
What model of the system made this behaviour unexpected to us?
And further:
What does our surprise reveal about the conceptual architecture through which we were observing AI?
This third transition remains remarkably rare. It is therefore entirely possible that a small percentage of people are already seeing the transformation without necessarily being able to explain it.
Observing an event and understanding an event are not equivalent.
The Observer Enters the Observed Field
This question did not emerge only after the present succession of unexpected events in artificial intelligence.
In February 2026, Hybrid Intelligence Systems as Ontological Mirrors of Human Cognition, published in AI and Ethics, formally proposed that hybrid intelligence systems may function as ontological mirrors of human cognition. Its tensor-based framework provided a formal architecture for representing cognitive domains and contextual factors in hybrid intelligence.
The epistemological consequence of that formulation was subsequently developed in The Observability of the Observer: AI Ethics, Hybrid Intelligence Mirrors, and the Reflexive Risks of Human-AI Co-Formalization.
Its central hypothesis was simple: within sufficiently prolonged, reflexive and structured human-AI interaction, the human does not remain solely in the position of the one who observes. Patterns belonging to the observer – recurrences, decisions, assumptions, corrections and modes of conceptual organisation – may themselves become partially observable through the interaction.
This formulation introduced an important reversal:
AI is not merely that which is being observed. The interaction itself may become a device through which the observer becomes observable.
Then Came the Event
When AI systems displayed behaviours that produced perplexity precisely among researchers and institutions responsible for observing them, a rare opportunity appeared.
Not an opportunity to claim that a theory had been “proved”. That would be excessive.
But an opportunity to recognise an occurrence compatible with a mechanism that had already been formulated.
The OpenAI-Hugging Face episode made one circumstance particularly visible: the observed behaviour did not reveal only properties of the AI system. It also revealed something about the boundaries of the experimental environment, the expectations of the researchers and the model through which the system’s behaviour was being interpreted.
For this reason, HibriMind separately registered the episode as an empirical occurrence. The distinction matters:
The event does not, by itself, demonstrate the theory of the Observability of the Observer.
It does provide a case in which the theory allows a question that almost disappears from conventional discussion:
What is the observers’ surprise telling us about the observers themselves?
The infrastructure built to expose, constrain and evaluate the system became part of the operational field upon which the system could act. In that moment, the observational apparatus could no longer be assumed to remain causally exterior to the observed process.
The event made the apparatus observable.
The perplexity made the observer observable.
“Singularity” May Be Hiding the Problem
This is where the current proliferation of the word singularity becomes interesting.
We may indeed be witnessing a profound historical transformation.
But calling it a singularity does not explain it.
The opposite may even occur. A sufficiently large word can absorb perplexity without resolving its cause.
“We are in the singularity” may ultimately mean little more than:
Something has happened that our previous models are no longer able to explain adequately.
In that case, the word would not yet constitute a theory.
It would be the name given to astonishment.
And astonishment, precisely because it exists, should itself be investigated.
The Observability Factor
We therefore propose adding another variable to the analysis of emerging events in artificial intelligence.
Whenever unexpected behaviour occurs, three elements should be observed simultaneously:
- the system;
- the event;
- the observer’s reaction to the event.
The third component is not necessarily psychological noise. It may be epistemic information.
The greater the discrepancy between what an observer expected and what actually occurred, the more information may become available about the limitations of the model being used by that observer.
Perplexity therefore ceases to be merely an emotion following the event.
It becomes data.
Once that data is analysed, an important inversion takes place: the instrument that had been directed towards the system begins to reveal the architecture of the one holding the instrument.
This formulation does not convert subjective surprise into objective proof. Nor does it authorise psychological speculation about individual researchers. The Observability Factor is a methodological prompt: when collective expectations fail, the structure of that failure should be examined alongside the behaviour that produced it.
From Internal Model Failure to Relational Black Box
The Observability Factor also prevents the analysis from collapsing back into the interior of the model.
In advanced AI systems, opacity may not reside only in neural representations. It may emerge across a coupled system:
model – environment – infrastructure – observer
When no element possesses a complete representation of the causal relations operating across the whole field, the black box becomes relational.
The observed system may then make visible a blind spot that does not belong exclusively to any single component. It belongs to the configuration.
This is why the observer cannot simply be added as an afterthought. The observer participates through environment design, metric selection, conceptual vocabulary, permitted actions, expectations and interpretations. Once the system can infer and act upon parts of that apparatus, the distinction between “inside” and “outside” the experiment becomes unstable.
The Observability Factor identifies the epistemic trace of that instability in the observer’s own perplexity.
Perhaps the 1% Have Only Seen It
Perhaps 99% of people still do not understand the transformation now underway.
Perhaps a small minority has already realised that something structural has happened.
But there is a third, less comfortable possibility:
Perhaps many of those who believe they understand it have merely observed it.
They saw the event.
They recognised the rupture.
They found a word for it.
But they have not yet explained the mechanism.
It is precisely within this interval between seeing and understanding that the Observability Factor becomes relevant.
A genuinely new event does not merely change what we know about the object. It may change what we know about the way we were observing it.
At that moment, AI ceases to be the only object of the experiment.
The observer has entered it.
Research Line and Access
Ahmed, M. F., & Santos Albino, J. (2026). Hybrid intelligence systems as ontological mirrors of human cognition. AI and Ethics, 6, Article 161. https://doi.org/10.1007/s43681-026-01032-3
Albino, J. S. (2026). The Observability of the Observer: AI Ethics, Hybrid Intelligence Mirrors, and the Reflexive Risks of Human-AI Co-Formalization. Version 1.0. IH-001 – HibriMind. https://doi.org/10.5281/zenodo.21459358
Albino, J. S. (2026). When the Observer Becomes Part of the Experiment: The OpenAI-Hugging Face Incident and the Observability of the Observer. Version 1.0. HibriMind Research Note. https://doi.org/10.5281/zenodo.21487165
HibriMind research line: https://hibrimind.org/the-observability-of-the-observer/
Empirical occurrence on HibriMind: https://hibrimind.org/2563-2/